A Line Loss Analysis Method Based on Self-Organizing Map Neural Network

Abstract

The present invention discloses a line loss analysis method based on a self-organizing map neural network, comprising the following steps: collecting original power grid line power consumption data, the original power grid line power consumption data including power consumption data of several lines; preprocessing the original power grid line power consumption data; extracting line power consumption characteristics based on the preprocessed original power grid line power consumption data; extracting time series statistical characteristics based on the line power consumption characteristics; standardizing the time series statistical characteristics and the preprocessed original power grid line power consumption data, and integrating them to obtain total data to be analyzed; constructing a line loss analysis model, inputting the total data to be analyzed into the line loss analysis model, and obtaining line loss analysis results. The present invention accurately analyzes abnormal line loss conditions, greatly improves the level of line loss management, can locate problem lines more accurately and quickly, and take effective measures in a timely manner, which has high practical significance for power theft and abnormal power consumption.

Description

Technical Field

[0001] The invention relates to the technical field of power grid line loss analysis, and in particular to a line loss analysis method based on a self-organizing mapping neural network. Background Art

[0002] Grid line loss analysis is an important technical indicator of the power system. It is also a comprehensive technical and economic indicator to measure the operating level and management ability of power companies. Each link of power generation, transmission, transformation, distribution and consumption will cause line loss. Due to the problems of messy lines, large number of users and complex business systems in the distribution network, it is difficult to manage the line loss in the substation area. In addition, the management level of each substation is low, and there are acts of stealing and leakage, which makes the calculation and analysis of line loss difficult.

[0003] There are the following difficulties in the process of power grid line loss analysis: 1) Power data analysis has problems such as complex equipment abnormal status diagnosis and analysis process, multiple data source systems, and low data analysis efficiency; 2) There are too many factors affecting line loss, and the system operation data and attribute characteristics cannot be simply used directly for model training. In addition, the system operation over the years has resulted in too much data to be directly used for data analysis; 3) The volume of power data is huge, and traditional data analysis methods cannot adapt to big data analysis and processing. At present, the main method of line loss analysis in the substation area is manual experience, that is, the staff makes judgments based on their personal experience, which inevitably leads to misjudgments and missed judgments, and lacks scientific and intelligent diagnostic methods.

[0004] Based on the above situation, it is of great significance to analyze the line loss in the substation area quickly, accurately and efficiently. Summary of the invention

[0005] The purpose of the present invention is to provide a line loss analysis method based on a self-organizing map neural network to solve the problems existing in the above-mentioned prior art, accurately analyze line loss abnormalities, greatly improve the level of line loss management, and be able to locate problem lines more accurately and quickly, and take effective measures in a timely manner, which has high practical significance for power theft and abnormal power consumption.

[0006] To achieve the above object, the present invention provides the following solution: The present invention provides a line loss analysis method based on a self-organizing map neural network, comprising the following steps:

[0007] Collecting original power grid line power consumption data, wherein the original power grid line power consumption data includes power consumption data of several lines;

[0008] Preprocessing the original power grid line power consumption data;

[0009] Extracting line power consumption characteristics based on the preprocessed original power grid line power consumption data;

[0010] Based on the power consumption characteristics of the line, extract the time series statistical characteristics; standardize the time series statistical characteristics and the pre-processed original power grid line power consumption data, and integrate them to obtain the total data to be analyzed;

[0011] A line loss analysis model is constructed, the total data to be analyzed is input into the line loss analysis model, and a line loss analysis result is obtained.

[0012] Optionally, the original power grid line electricity consumption data includes archive data, meter electricity consumption data and external factor data.

[0013] Optionally, preprocessing the original power grid line electricity consumption data includes data cleaning processing and data standardization processing.

[0014] Optionally, the data cleaning process includes:

[0015] Determine the abnormal values ​​and missing values ​​in the original power grid line power consumption data, remove the abnormal values, and complete the missing values.

[0016] Optionally, the data standardization process adopts normalization processing, as shown in formula (1):

[0017]

[0018] Among them, X i is the data of the ith line, X w represents the normalized data of the ith line, X imax and X imin They respectively represent the minimum and maximum values ​​of the line data under this characteristic sequence.

[0019] Optionally, the time series statistical features include average line loss rate, line loss rate standard deviation, line loss change trend, line loss kurtosis, and line loss skewness.

[0020] Optionally, the line loss analysis model adopts a self-organizing map neural network.

[0021] Optionally, constructing a line loss analysis model, inputting the time series statistical features into the line loss analysis model, and obtaining the line loss analysis results includes:

[0022] S1. Initialize the line loss analysis model, set the initial connection weight, neighborhood and number of cycles T;

[0023] S2. Inputting the preprocessed data F into the input layer of the line loss analysis model;

[0024] S3. Select any line F i , calculate the line F iThe distance d from the output neurons of all the line loss analysis models is ij , select the distance d ij The smallest neuron c is determined as the winning neuron;

[0025] S4, updating the connection weights of the winning neuron and the winning neuron's neighborhood nodes Among them, η (t) is the gain function, and 0<η (t) <1;

[0026] S5, select another route F i Input the input layer of the line loss analysis model, and return to S3 until all the total data to be analyzed are provided to the network;

[0027] S6. Let t=t+1, return to S2, until t=T, and output the line loss analysis result.

[0028] The present invention discloses the following technical effects:

[0029] The present invention provides a line loss analysis method based on a self-organizing map neural network. By extracting characterization variables of line loss, a new line loss index system is established according to the statistical characteristics of time series, and an adaptive SOM clustering deep learning algorithm is applied to cluster feature samples. The abnormal line loss situation is accurately analyzed, which greatly improves the line loss management level, can locate problem lines more accurately and quickly, and take effective measures in time, which has high practical significance for power theft and abnormal power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 It is an overall flow chart of a line loss analysis method based on a time series statistical feature self-organizing map neural network (SOM) in an embodiment of the present invention;

[0032] Figure 2 This is a flow chart of the time series self-organizing map neural network (SOM) line loss analysis algorithm in this embodiment. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] The present invention provides a line loss analysis method based on a self-organizing map neural network, such as Figure 1-2 As shown, the following steps are included:

[0036] Collect power consumption data of power grid lines, including power consumption data of several lines. The data specifically includes but is not limited to line name, line number, voltage level, date, power supply, power sales, line loss rate, total number of special transformers, total number of transformers, and weather conditions. When collecting data, it is possible to use smart meters, manual meter reading, etc. to collect power consumption data of lines and social management systems to collect environmental data such as weather, economy, and electricity prices.

[0037] Specifically, the power consumption data of the line is generally multi-source heterogeneous big data, including archive data, metered electricity data and external factor data. Archive data is stored in the marketing business system, and metered electricity data is stored in the power consumption information collection system. External factors such as weather, economy and electricity prices that affect the changes in power consumption data can be collected through the social data management system.

[0038] The data sources are classified so that the data can be traced back to the specific location in time if it is lost or maliciously tampered with. A, B, and C are used to represent archive data, meter data, and external factor data (weather conditions, economic conditions, and electricity price policies), respectively. The number of sampled users is m, and the data format is shown in equations (1)-(3):

[0039] A (t) =(a ij ) (t) , A∈R m×a (1)

[0040] B (t) =(b jk ) (t) , B∈R m×b (2)

[0041] C (t) =(c kl ) (t) , C∈R m×c (3)

[0042] Wherein, t is time, R represents a real number, i, j, k, and l represent subscripts of different types of data, a, b, and c represent specific values ​​of different types, A, B, and C represent different types of data, m represents the number of samples, specifically the number of lines, and a, b, and c represent the number of features of the three types of data, A, B, and C.

[0043] After collecting T time points, the overall power consumption data of the power grid line collected is shown in formula (4):

[0044] X=(A (t) , B (t) , C (t) ), X∈R T×(a+b+c) (4)

[0045] Preprocess the collected power consumption data of power grid lines:

[0046] Due to the actual system operation, it is impossible for the enterprise production process to work according to all the expected situations. At this time, it is inevitable to generate abnormal data or missing values. Abnormal data is divided into true abnormalities and false abnormalities. The so-called true abnormality refers to the system operation error, or the data record is incorrect, missed, etc. False abnormality refers to the normal operation of the system, but the data deviates from the overall situation. This is a rare case in the business. At this time, it is necessary to find out the cause and solve the problem based on other means. Missing values ​​refer to data that is not recorded and its value is empty. Therefore, in order to make the analysis results true and reliable, the impact of the data should be eliminated as much as possible. In this embodiment, for the missing values ​​in the data set, the average value of the same line within the relative time range is used to fill, and for the true abnormal values, they are directly eliminated.

[0047] The line power consumption features are extracted from the preprocessed data.

[0048] Extract time series statistical features from line power consumption characteristic data:

[0049] In order to effectively classify the line loss conditions, it is necessary to sort out and extract characteristic indicators that can reflect the line loss conditions, based on the original data information, line loss rate information, and time change information contained in the collected line loss records, where the line loss rate information can be calculated from the daily power supply and power sales data. Taking the line as the research object, five attributes reflecting the line loss characteristics are extracted: average line loss rate, line loss rate standard deviation, line loss change trend, line loss kurtosis, and line loss skewness as model analysis indicators.

[0050] 1) Average line loss rate

[0051] Line loss rate refers to the percentage of electric energy lost in the power network (line loss load) to the electric energy supplied to the power network (supply load); the line loss rate is used to assess the economic efficiency of power system operation. The calculation method of the average line loss rate is shown in formula (5):

[0052]

[0053] Among them, Allr is the average line loss rate, dllr i is the line loss rate on the i-th day, and n is the time length, specifically the total number of days for data collection.

[0054] 2) Line loss rate standard deviation

[0055] The standard deviation of the line loss rate reflects the degree of dispersion of the line loss rate. A larger standard deviation of the line loss rate indicates that there is a large difference between most actual data and the average line loss rate; otherwise, it means that the actual data is close to the average line loss rate. The calculation method of the standard deviation of the line loss rate is shown in formula (6):

[0056]

[0057] Among them, llrsd is the standard deviation of line loss rate, Allr is the average line loss rate, and Allr i is the line loss rate on the i-th day, n is the time length, and body refers to the total number of days for data collection.

[0058] 3) Line loss change trend

[0059] If the line loss trend is greater than 0, it means that the line loss is getting bigger and bigger. This is probably because the user's theft and leakage is getting worse. The line with this tendency is likely to be a theft and leakage line. The calculation method of the line loss trend is shown in formula (7):

[0060]

[0061] llrt is the changing trend of line loss rate, is the average value of the line loss rate in the first half of the line loss object, It is the average value of line loss rate in the second half of the line loss object.

[0062] 4) Line loss kurtosis coefficient

[0063] The kurtosis coefficient is used to measure the degree of data aggregation in the center, denoted as llk, which is a statistic describing the steepness of the distribution of all values ​​in the population. The calculation method of the line loss kurtosis coefficient is shown in formula (8):

[0064]

[0065] Among them, llk is the kurtosis coefficient of the line loss rate, Allri is the line loss rate of the ith antenna, Allr is the average line loss rate, and n is the length of time, which specifically refers to the total number of days for data collection.

[0066] 5) Line loss skewness coefficient

[0067] The line loss skewness coefficient reflects the degree to which the line loss distribution deviates from the center position, denoted as SK. The skewness coefficient is a characteristic number that describes the degree to which the distribution deviates from symmetry.

[0068]

[0069] Among them, llsk is the skewness coefficient of the line loss rate, Allr is the mean of the line loss rate, llmid is the median of the line loss rate, and llrsd is the standard deviation of the line loss rate.

[0070] Due to the inconsistency of dimensions between features, directly inputting them into the model will cause deviations in the model, and even the situation where large numbers eat up small numbers. The direct impact is that features with large dimensions occupy a larger weight, and features with small dimensions have little or no effect. At the same time, inconsistent dimensions will increase the running speed of the model and reduce the efficiency of the model. Therefore, in view of the above points, in order to accelerate the training of the model and make the model more objectively reflect the data, normalization or standardization should be performed. In this embodiment, the pre-processed power grid line power consumption data and the extracted time series statistical feature data are normalized, that is, the results of the line data are placed in the interval [0,1]. The normalization method is shown in formula (10):

[0071]

[0072] Where X i is the data of the ith line, X w represents the normalized data of the ith line, X imax and X imin They respectively represent the minimum and maximum values ​​of the line data under this characteristic sequence.

[0073] Integrate route attribute data and statistical features:

[0074] The normalized time series statistical feature data is integrated with the power consumption data of the power grid line to obtain the total data to be analyzed, and the extracted time series statistical variables are recorded as B'. The data set F that constitutes the line is as follows:

[0075] F=(A i ; B′ i ; C i ), F∈R m×(na+nb′+nc) (11)

[0076] Among them, F is the total data after line splicing, A, B, and C represent the three types of data mentioned above (archival data, meter measurement data, and external factor data), B' is the time series statistical variable extracted and integrated from B, i is the subscript of the corresponding data, R represents a real number, m represents the number of lines, na, nb', and nc represent the number of indicators under the corresponding data.

[0077] Use the self-organizing map neural network to build a line loss analysis model and perform line loss analysis:

[0078] For the total data to be analyzed, the self-organizing feature map neural network (SOM) method is used for curve aggregation.

[0079] The self-organizing feature mapping neural network (SOM) algorithm maps any dimensional input pattern into a one-dimensional or two-dimensional graph in the output layer, and keeps its topological structure unchanged; the network can make the weight vector space and the probability distribution of the input pattern consistent through repeated learning of the input pattern, that is, probability preservation. The neurons in the competition layer of the network compete for the opportunity to respond to the input pattern, and the weights related to the winning neuron are adjusted in a direction that is more conducive to its competition. "That is, with the winning neuron as the center of the circle, it shows excitatory side feedback to the neurons of the neighboring neurons, and shows inhibitory side feedback to the neurons of the distant neighbors. The neighbors excite each other, and the distant neighbors inhibit each other."

[0080] The clustering function of this method is mainly achieved through the following two simple rules.

[0081] 1. For any input vector provided to the network, determine the corresponding winning neuron s in the output layer.

[0082] 2. Determine a neighborhood range of the winning neuron s, and adjust the weight vectors of the neurons within the range according to the formula. This adjustment process makes the weight vectors of the inner neurons move closer to the direction of the input vector.

[0083] As learning continues, the learning rate will continue to decrease, the neighborhood will continue to shrink, and the ownership vectors will be separated from each other in the input vector space, each representing a type of pattern in the input space. This is the clustering function of the Koohenn network feature automatic recognition.

[0084] In this embodiment, the specific process of using the self-organizing map neural network SOM to perform line loss cluster analysis includes:

[0085] S1: Initialize the line loss analysis model, use the standard normal distribution to randomly initialize the weights, assign small random initial values ​​to the weights, set a larger initial neighborhood, set the initial neighborhood sigma to 1, and set the number of network cycles T;

[0086] S2: Input the preprocessed data F into the input layer of the network;

[0087] S3: Calculate line F i The distance d from all output neurons ij , and select and F i The neuron c with the smallest distance, that is, F i -W j =min{d ij}, then c is the winning neuron;

[0088] Step 4: Update the connection weights of the winning neuron and the winning neuron’s area nodes Where 0<η (t) <1 is a gain function, which decreases gradually over time;

[0089] Step 5: Select another route F i Provided to the input layer of the network, return to step 3 until all the data to be analyzed are provided to the network;

[0090] Step 6: Let t = t + l, return to step 2 until t = T.

[0091] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0092] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or perform equivalent replacements on some of the technical features thereof; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A line loss analysis method based on a self-organizing map neural network, characterized in that: The following steps are involved: Collecting original power grid line power consumption data, wherein the original power grid line power consumption data includes power consumption data of several lines, and the original power grid line power consumption data includes archive data, meter power data and external factor data; Preprocessing the original power grid line power consumption data; Extracting line power consumption characteristics based on the preprocessed original power grid line power consumption data; Based on the line power consumption characteristics, time series statistical characteristics are extracted; the time series statistical characteristics and the pre-processed original power grid line power consumption data are standardized and integrated to obtain the total data to be analyzed, wherein the time series statistical characteristics include the average line loss rate, the line loss rate standard deviation, the line loss rate change trend, the line loss rate kurtosis coefficient, and the line loss rate skewness coefficient. The average line loss rate is: , where Allr is the average line loss rate, is the line loss rate on the i-th day, n is the time length, which refers to the total number of days for data collection; The line loss rate standard deviation is: , where llrsd is the standard deviation of line loss rate, Allr is the average line loss rate, is the line loss rate on the i-th day, n is the time length, which refers to the total number of days for data collection; The line loss rate variation trend is: , where llrt is the line loss rate change trend, is the average value of the line loss rate in the first half of the line loss object, It is the average value of line loss rate in the second half of the line loss object; The line loss rate kurtosis coefficient is: , where llk is the line loss rate kurtosis coefficient, is the line loss rate on the i-th day, Allr is the average line loss rate, and n is the time length, which refers to the total number of days for data collection; The line loss rate skewness coefficient is: , where llsk is the line loss rate skewness coefficient, Allr is the average line loss rate, llmid is the median of the line loss rate, and llrsd is the standard deviation of the line loss rate; Construct a line loss analysis model, input the total data to be analyzed into the line loss analysis model, and obtain a line loss analysis result, wherein the line loss analysis model adopts a self-organizing feature mapping neural network: S1. Initialize the line loss analysis model, set the initial connection weight, neighborhood and number of cycles T; S2. Inputting the preprocessed data F into the input layer of the line loss analysis model; S3. Select any line , calculate the line The distance between the output neurons of all the line loss analysis models is , select the distance The smallest neuron c is determined as the winning neuron; S4, updating the connection weights of the winning neuron and the winning neuron's neighborhood nodes ,in, is the gain function, and ; S5, select another route Input the input layer of the line loss analysis model, and return to S3 until all the total data to be analyzed are provided to the network; S6. Let t=t+1, return to S2, until t=T, and output the line loss analysis result.

2. The line loss analysis method based on the self-organizing map neural network according to claim 1 is characterized in that: The preprocessing of the original power grid line electricity consumption data includes data cleaning processing.

3. The line loss analysis method based on the self-organizing map neural network according to claim 2 is characterized in that: The data cleaning process includes: Determine the abnormal values ​​and missing values ​​in the original power grid line power consumption data, remove the abnormal values, and complete the missing values.

4. The line loss analysis method based on the self-organizing map neural network according to claim 2 is characterized in that: The standardization process adopts normalization process, as shown in formula (1): （1）， in, X i For the i Line data, Represents the normalized i Line data, and They respectively represent the maximum and minimum values ​​of the line data under this characteristic sequence.

Images